Detect and handle mass-downvoting

[michaeldickens]

I noticed that Diego Caleiro appears to be getting mass-downvoted by someone (a lot of his comments are at -1). I’d consider this bad behavior. There could be a system to detect if person A repeatedly downvotes person B in a short period of time, and disregards their downvotes. I believe reddit works this way.

[tog22]

I agree that that's bad behaviour, and have occasionally noticed similar pattens on the forum. However we probably ought to do some community consultation before implementing such a feature and then warn people, since many forum readers presumably assume their downvotes are private. What do you think @ryancarey? Should you or @michaeldickens post about this on the latest open thread, or on a separate post on the forum?

[patbl]

This has been a problem on Less Wrong:

http://lesswrong.com/lw/77b/please_do_not_downvote_every_comment_or_post/ http://lesswrong.com/lw/jo7/a_few_remarks_about_massdownvoting/ http://lesswrong.com/lw/kbk/meta_policy_for_dealing_with_users/

Maybe there's something to be learned there. Dunno what they ended up doing.

[Discordius]

Yeah, LessWrong dealt with this a good bit. Their general solution was to have mods analyze the logs for downvoting (by messaging TrikeApps) and then be fairly harsh in banning or punishing the relevant offenders. I think people have a reasonable expectation of their voting behavior to be analyzed by the mods, since upvoting and downvoting rings are a pretty common problem on the internet, and people know that abuse will normally be punished (e.g. Hacker News, Reddit and most other similar places all have that kind of moderation in place)

I think making an announcement before acting on this policy seems sensible, but I would also encourage at least warning past offenders, and if the offense was bad enough, make that warning public.

I can ping Robby or Kaj Sotala for concrete input on their experiences at LessWrong, they might have some wisdom to add.

[tog22]

I can ping Robby or Kaj Sotala for concrete input on their experiences at LessWrong, they might have some wisdom to add.

Go for it!

[peterhurford]

@Discordius @tog22 @patbl @michaeldickens Is there a tractable solution to this problem? Do we have a path forward?

[michaeldickens]

I don't really know but I have an idea of how to implement this. When someone downvotes a post/comment, check their recent history of voting on this user. If all their recent votes are downvotes, don't count any of them.

[peterhurford]

@michaeldickens - Maybe I'd rephrase that to "When someone downvotes a post/comment, check their recent history of voting on this user within the past day. If all their recent votes are downvotes, delete all their votes for this user within the past day."

[patbl]

Here are some questions:

When mass downvoting is detected, should the downvote button be visible? Should it be disabled and have an explanatory tooltip? Another possibility would be to allow the UI to remain as it is but avoid recording downvotes in the database. (If a mass downvoter refreshed the page, the downvote button would become clickable again.)

Should someone be able to downvote without restriction if they upvoted even once within the past day? Maybe there should be a hard limit (20 downvotes per day, regardless of the number of upvotes). There could also or instead be a ratio requirement (at most two downvotes for every upvote). It seems that some combination of the two might be best, but it would be more difficult to explain to users.

It might be worth trying to find out how Stack Exchange sites or Reddit handle this.

It could also be helpful to examine the voting patterns of well-behaved users so that we don't annoy them with new restrictions. But I doubt whether such information is easily obtained.

[RyanCarey]

Just avoid accidentally blocking me / other good-spirited users with this one!

Often someone will post a handful of unhelpful comments and I'll be scanning through recent comments and downvote them all!

So I don't think >=3 downvotes with >90% negativity would be a sufficiently specific filter... Maybe it works if you use a higher number, and only count posts that are >1 day old?

Ryan Carey | bioinformatics student | medical resident

On Wed, Apr 27, 2016 at 10:11 AM, Peter Hurford notifications@github.com wrote:

@michaeldickens https://github.com/michaeldickens - Maybe I'd rephrase that to "When someone downvotes a post/comment, check their recent history of voting on this user within the past day. If all their recent votes are downvotes, delete all their votes for this user within the past day ."

— You are receiving this because you were mentioned. Reply to this email directly or view it on GitHub https://github.com/tog22/eaforum/issues/47#issuecomment-214926329

[patbl]

I don't have access to the production database, so I can't test algorithms against real-world data to see how successful they are at accurately identifying inappropriate downvoting patterns.

The manual approach might work. We could build an admin page that shows a user's voting history. People would be banned only in response to complaints or suspicious patterns happened upon by admins. Banning is pretty harsh, though, and I don't know what form of punishment is available short of it.

The only restriction on downvoting is that it costs you one point per downvote. Maybe we should require some number of points (greater than one) before allowing downvoting. Stack Exchange does this, and the threshold is fairly high (125 points). If only high-rep people had access to downvoting, inappropriate uses of it would be much lessened. The trade-off would be less appropriate downvoting.

Ultimately, though, points are a fiction.

[RyanCarey]

Costing one point to vote seems like a good status quo. It means you can't pump votes by making lots of accounts, unless you had a lot already, right?

I'm not to fazed by this, and you can do what you think is needed.

On Oct 7, 2016 9:46 PM, "Patrick Brinich-Langlois" notifications@github.com wrote:

I don't have access to the production database, so I can't test algorithms against real-world data to see how successful they are at accurately identifying inappropriate downvoting patterns.

The manual approach might work. We could build an admin page that shows a user's voting history. People would be banned only in response to complaints or suspicious patterns happened upon by admins. Banning is pretty harsh, though, and I don't know what form of punishment is available short of it.

The only restriction on downvoting is that it costs you one point per downvote. Maybe we should require some number of points (greater than one) before allowing downvoting. Stack Exchange does this, and the threshold is fairly high (125 points http://stackoverflow.com/help/privileges/vote-down). If only high-rep people had access to downvoting, inappropriate uses of it would be much lessened. The trade-off would be less appropriate downvoting.

Ultimately, though, points are a fiction.

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/tog22/eaforum/issues/47#issuecomment-252402699, or mute the thread https://github.com/notifications/unsubscribe-auth/AH9Fu-Noh37ybmjsMc4St4Sw0f845T8Uks5qxyAogaJpZM4Gsh_X .

[tog22]

Rereading our exchanges above I have a suspicion that there's no very good, tractable, likely-to-get-implemented solution which is particularly superior to the status quo. Patrick's idea of requiring more points to downvote (I'd say 15) seems the simplest improvement. I'd suggest Patrick say whether there's anything he thinks he should and would do, people then have 5 days to comment, and unless that changes things he then goes ahead or closes this issue as appropriate. 😄

[RyanCarey]

Oh wait, you're required to have points for downvoting, not for upvoting. I see. this means people can still pump votes by creating accounts.

I don't think increasing the required amount of votes for downvoting helps

• it's the upvoting that seems more relevant to me, because people use that for self-promotion.

So of the suggestions we have so far, I'd favor the status quo.

On Fri, Oct 7, 2016 at 10:54 PM, tog22 notifications@github.com wrote:

Rereading our exchanges above I have a suspicion that there's no very good, tractable, likely-to-get-implemented solution which is particularly superior to the status quo. Patrick's idea of requiring more points to downvote (I'd say 15) seems the simplest improvement. I'd suggest Patrick say whether there's anything he thinks he should and would do, people then have 5 days to comment, and unless that changes things he then goes ahead or closes this issue as appropriate. 😄

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/tog22/eaforum/issues/47#issuecomment-252405145, or mute the thread https://github.com/notifications/unsubscribe-auth/AH9Fu7rjRW4m9AVPxY-VPmIO8V5NzIiAks5qxzAKgaJpZM4Gsh_X .

Ryan Carey Assistant Research Fellow Machine Intelligence Research Institute

[peterhurford]

I'm okay with punting on this.